Downloads · 30 days
0
michaellu88/mnn-engines
mnn-engines is a machine learning model from michaellu88. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
The eight MNN runtime libraries used by the Pashbox Books Android app (native/modules/mnn-vision-bridge). Published publicly so the app and its build tooling can fetch them without an access token.
Downloads · 30 days
0
Access
Public
Updated Sep 23, 2026
Repo size
7.5 MB
Likes
0
Public
Click a slice to open those files.
.so7.6 MB · 100%
From the Hugging Face model README
The eight MNN runtime libraries used by the Pashbox Books
Android app (native/modules/mnn-vision-bridge). Published publicly so the app and its
build tooling can fetch them without an access token.
These are the MNN inference engine only. No model weights are here — the trained OCR models are kept in a separate private repository.
| File | Bytes | sha256 |
|---|---|---|
arm64-v8a/libMNN.so | 2,398,200 | 69d6db5ade18139b58d2d96430b741f18b9a708083947f75815e08f36ee17ef2 |
arm64-v8a/libMNN_CL.so | 2,135,960 | 06388cb10b59e3d3f0b568e56b587d74c50a4a5abba639d58af8e92b922f0f19 |
arm64-v8a/libllm.so | 1,206,232 | 7d50559f48eb23a7d517d03147c430aee539af15712aa08e09531f2923ade0b5 |
arm64-v8a/libMNN_Express.so | 743,896 | 0cc0c9fca8426c55cd795f619ae579a65356dbee75a700ceb5937781c63ec5b4 |
arm64-v8a/libMNN_Vulkan.so | 738,600 | 43eb831d7c71de3a819873bf0eabda79c3b67a6b463786133ce5a511516027c0 |
arm64-v8a/libMNNOpenCV.so | 264,424 | a18f299abb65679bc71213b534fbcc843c498c90b867e4f16df6edd31495469e |
arm64-v8a/libMNNAudio.so | 64,248 | d23fcf526aa78b64f105cfc622d53f88c51a7a9103687f4189c1d28ec3faddd5 |
arm64-v8a/libmnncore.so | 22,816 | 40705e8a12b6ca6df18557536f4013d7b05069dbfbdccd17b4f8925feacaa8dd |
libMNN.so carries the build paths of the upstream alibaba/MNN GitHub Actions CI
(/home/runner/work/MNN/MNN/build_64/...), and bundles Arm's
KleidiAI 1.14.0 microkernels. A string scan of all
eight libraries found no credentials, no personal paths and no internal hostnames.
Read from each library's DT_NEEDED entries — this is the real dependency graph, not a
listing order:
libMNN.so -> libc++_shared.so
libMNN_Express.so libMNN_CL.so
libMNN_Vulkan.so libmnncore.so -> libMNN.so
libMNNAudio.so libMNNOpenCV.so -> libMNN_Express.so, libMNN.so
libllm.so -> libMNN_Vulkan.so, libMNN_CL.so, libMNNOpenCV.so,
libMNNAudio.so, libMNN_Express.so, libMNN.so
A valid load order is therefore: libMNN → libMNN_Express, libMNN_CL,
libMNN_Vulkan, libmnncore → libMNNAudio, libMNNOpenCV → libllm.
Two things are not in this repo:
libc++_shared.so — every library here needs it. It is the Android NDK C++ runtime
and ships inside the APK; it is not redistributed here.libmnn_bridge.so — the app's JNI bridge. It is built from mnn_bridge.cpp by CMake
during the Android build and is never downloadable.Place the eight files in modules/mnn-vision-bridge/android/src/main/jniLibs/arm64-v8a/
before building. The APK then ships them and MnnVisionBridgeModule loads them at startup,
where the dynamic linker resolves every DT_NEEDED dependency from the app's native
library directory.
Loading them at runtime with System.load() from an arbitrary app-data path is not
equivalent: that path is not on the linker's search path, so dependencies are resolved from
elsewhere or not at all, and libc++_shared.so is not here to be downloaded.
MNN is licensed under the Apache License 2.0 — see LICENSE, copied verbatim from
alibaba/MNN. KleidiAI is likewise
Apache-2.0, © Arm Limited. This repository redistributes those binaries; it is not
affiliated with or endorsed by Alibaba or Arm.